DEX vs CEX Volume: The Record Share Hides a Two-Year Low

A ratio can set a record while both of the numbers inside it are shrinking.

In July 2026, spot volume on decentralized exchanges came to 24.14% of spot volume on centralized ones. That is the highest monthly reading since the series began in 2019, and for much of 2024 the same measure sat below 10%. The figure is real, it is verifiable, and it arrived attached to a tidy story about order flow moving on-chain for good. From the same dataset, in the same month, absolute DEX volume fell roughly 26% to about $130.77 billion, the thinnest month in nearly two years.

Both books shrank. One shrank less.

This walkthrough follows Eunha, who lives between structure and emotion and takes an argument apart by asking rather than asserting, and Nina, twenty-three, eleven months into her first brokerage account, who checks numbers against their source for sport. Nina arrives already holding the headline, and already certain what it means.

"A quarter of spot trading is on-chain now," she says, turning her phone around. "That is the whole thesis, finally proven."

"That is one reading of it," Eunha says. "Before you take it, tell me what the two numbers are."

What the 24% actually divides

Nina glances back at the chart. "DEX volume over CEX volume."

"Where does each of them come from?"

There is a pause, and it is the useful kind. The ratio is quoted so often that the question underneath it stops getting asked: not what the number is, but what produced it.

Centralized exchange volume is reported by the exchange. A trade matches in an order book, the venue records the size, and the venue publishes the total. Nobody outside the company observes the match directly. On-chain volume works the other way. DefiLlama builds it by reading swap events emitted by pool contracts, so the source is a public log rather than a company statement.

"So one of them is honest and one of them is a press release," Nina says.

"One of them is publicly verifiable," Eunha says. "That is not the same claim. Verifiable means nobody edited the log after the fact. It says nothing about what the log was counting."

The July reading sits inside a real trend. The Block frames it as 24%, up from around 17% a year earlier, and the direction has been consistent for well over a year. The activity is concentrated on a handful of chains: measured over a trailing 30-day window, Solana carried about $49.86 billion of spot DEX volume, BNB Chain about $31.04 billion, Ethereum about $28.84 billion, and Base about $22.38 billion. Those windows do not line up with the July calendar month. Add the four together, reconcile them against $130.77 billion, and the result means nothing.

Nina writes down the four chains. "Fine. The trend is real. So why are you being careful about it?"

"Because a rising percentage has two possible causes and the headline only names one."

Why did the DEX share hit a record while DEX volume fell?

The cause the headline does not name is a shrinking denominator.

A ratio goes up when the top gets bigger or when the bottom gets smaller. In July both terms moved down. On-chain spot volume dropped about 26% month over month to roughly $130.77 billion, the lowest in close to two years. Centralized spot volume dropped further. The record is a description of relative decline, and it was recorded during one of the quietest on-chain months since 2024.

The slower forces behind the trend are real, and they are worth separating from the month that just closed. Compliance requirements on centralized venues have tightened across the United States, Europe and several Asian jurisdictions, which raises the cost of listing anything early or unusual. Permissionless pools carry no such gate, so newly issued assets and the faster end of speculative activity tend to land there first. That explains a multi-year drift. It does not explain a single month, and July is a single month.

"That cannot be right," Nina says. "Every chart of this thing goes up and to the right."

"The chart is accurate. The chart is a fraction."

She sits with that. Then, more slowly: "So the share can hit an all-time high in a month where less trading happened everywhere."

"Yes. And a month where less trading happened on-chain specifically."

The discipline here is a habit, not a calculation. Before believing a ratio, find out which of its two terms moved. A record volume figure can be produced by a record built by one participant rather than by broad demand, and a record share can be produced by a collapse on the other side of the divide. Neither is a lie. Both are answers to a question nobody asked out loud.

Nina pushes back, and it is the right push. "But even if the share rose for a boring reason, the on-chain number is still the honest one. It settles publicly. You cannot fake it."

"Say more about what you mean by fake."

"Nobody can type a bigger number into a blockchain."

"Agreed," Eunha says. "Now tell me what a blockchain counts."

A matched trade prints once. A routed trade prints at every pool

An order book match is a single event with two sides. Someone lifts an offer, size changes hands, the venue records one trade. The unit is clean because the venue is the only place the trade exists.

On-chain, the unit is the swap event. A swap event belongs to a pool, not to an intention. Send one trade through three pools to get from the asset you hold to the asset you want, and each pool emits its own swap event. Each of those events feeds the chain's volume total. The economic intention was one trade. The measured surface was three.

"How often does that actually happen?" Nina asks.

"Often enough that the totals are not directly comparable," Eunha says, "and that is as far as the data will let anyone go."

The precision stops there on purpose. How individual adapters treat multi-hop routes inside a single router varies. Aggregators are tracked on separate dashboards with their own adapters, rather than stacked on top of the DEX total. There is no clean multiplier to apply, and anyone quoting one has invented it. What survives the uncertainty is the direction: on-chain volume counts execution steps, centralized volume counts matched trades. Divide the first by the second and the units do not cancel.

Nina is quiet for a moment. "So it is not that the number is dishonest. It is that it is answering a different question than I thought."

"Honest and comparable are separate properties, and only one of them is on offer here."

Routing is not an accounting quirk either. It shapes what you actually receive, which is why how a swap gets routed is worth understanding on its own terms before it shows up as a line in someone's monthly chart.

Two measurement regimes, one decimal point

There is a version of this argument that flatters on-chain markets, and it is the one that travels. Centralized venues self-report, self-reporting has a documented history of inflation, and public logs cannot be edited after the fact. All of that is true.

The uncomfortable half is that neither number is audited against the other. Self-reported volume has a wash-trading history. Event-log volume is faithful to what the contracts emitted and indifferent to how many of those emissions represent one person's single decision. There is no referee sitting between the two datasets confirming they describe the same thing, because they do not. Which is what makes 24.14% a strange figure to quote to two decimal places.

"It looks so precise," Nina says.

"It is precise," Eunha says. "Precision and accuracy are different words for a reason."

Where you draw the boundary changes the answer too. CoinGecko runs its own version of this series using the top ten DEXs against the top ten CEXs, and reads about 21.2% for November 2025. That is not a contradiction of the DefiLlama series and neither provider is wrong. They drew different boundaries around what counts, and the boundary is doing as much work as the market is.

Nina asks the question a beginner asks and an experienced desk forgets to. "Then which one do I use?"

"Whichever one you can describe. If you cannot say what a series includes, you cannot say what a change in it means."

That discipline generalizes past this ratio. The gap between an adjusted figure and a raw one is exactly which volume measure you are reading, and the answer changes what the same headline is allowed to claim.

What a share of volume does not tell you about your fill

Nina has been building toward this without saying it. "None of this changes what I do on Monday, does it?"

"Say what you were going to do on Monday."

"Take the on-chain venue. Twenty-four percent, growing, obviously deeper."

"Deeper than what, at what size?" Eunha asks.

Volume is a record of what has already happened. Depth is what is resting on the book right now, at the price you want, in the size you need. A venue can process enormous volume in small increments and still move against you the moment your order is larger than the pocket of liquidity in front of it. The ratio contains no information about that. It cannot: it is a monthly aggregate of completed activity, and your fill is a question about the next few seconds.

The absolute figure is closer to useful than the share, and even it is indirect. Roughly $130.77 billion moved on-chain in July against a much larger month in June. Less activity, spread across the same venues, on a measurement that counts hops. Whatever that says about adoption, it does not describe an environment where size clears more easily than it did before.

The concentration compounds it. Four chains carry the overwhelming share of on-chain spot activity, and inside each of them the depth sits in a small number of pools around a small number of pairs. An aggregate labelled "on-chain volume" is describing a set of venues where the pair you actually want can be thin in a month the total looks healthy. The ratio flattens all of that into one percentage, which is what a ratio is built to do and also why it cannot answer the question Nina brought to it.

"So the honest version is that the ratio tells me about the ratio."

"The honest version is that it tells you about two markets' relative direction, which is worth knowing, and nothing about execution, which is what you were about to use it for."

This is the same trap that sits inside open interest, where the number describes size against direction and gets read as a forecast. The metric is not broken. The question being asked of it is.

Nina closes the chart. "I still think the trend is real."

"It is. Nothing here says on-chain venues are not taking share over time. The claim that did not survive is that July proved it."

The distinction matters more than it looks. A structural migration and a faster contraction on one side produce the same line on the chart, and they imply completely different things about where the market is going. Telling them apart takes one extra step: look at the numerator and the denominator separately before reading the quotient. July gives both, from the same source, in the same month, and they point in opposite directions.

A record is only information once you know which term produced it.

Take the next ratio that shows up in your feed and find out which of its two numbers actually moved. Then open the simulator, put a position on something you have only ever read about as a percentage, and see how far a monthly share sits from the price that fills you. The $5,000 is simulated, which makes it a cheap place to learn the gap.

Start free →